DNA methylation is a key determinant of response to targeted and immune checkpoint therapies in metastatic renal cell carcinoma

DNA methylation is a key determinant of response to targeted and immune checkpoint therapies in metastatic renal cell carcinoma | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results DNA methylation is a key determinant of response to targeted and immune checkpoint therapies in metastatic renal cell carcinoma View ORCID Profile Florian Jeanneret , Sarah Schoch , Pedro Ballester , Stefan N Symeonides , Alexander Laird , View ORCID Profile Håkan Axelson , Delphine Pflieger , Christophe Battail doi: https://doi.org/10.1101/2024.11.05.622095 Florian Jeanneret 1 Université Grenoble Alpes, CEA, INSERM, UA13 BGE, CEA, CNRS , FR2048, Grenoble, France Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Florian Jeanneret Sarah Schoch 2 Division of Translational Cancer Research, Department of Laboratory Medicine Lund University ; Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Pedro Ballester 3 Department of Bioengineering, Imperial College London , London SW7 2AZ, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Stefan N Symeonides 4 Edinburgh Cancer Centre, Western General Hospital , NHS Lothian; Edinburgh, UK 5 Institute of Genetics and Cancer, Cancer Research UK Edinburgh Centre, University of Edinburgh ; Edinburgh, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Alexander Laird 5 Institute of Genetics and Cancer, Cancer Research UK Edinburgh Centre, University of Edinburgh ; Edinburgh, UK 6 Department of Urology, Western General Hospital , NHS Lothian; Edinburgh, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Håkan Axelson 2 Division of Translational Cancer Research, Department of Laboratory Medicine Lund University ; Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Håkan Axelson Delphine Pflieger 1 Université Grenoble Alpes, CEA, INSERM, UA13 BGE, CEA, CNRS , FR2048, Grenoble, France Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: christophe.battail{at}cea.fr delphine.pflieger{at}cea.fr Christophe Battail 1 Université Grenoble Alpes, CEA, INSERM, UA13 BGE, CEA, CNRS , FR2048, Grenoble, France Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: christophe.battail{at}cea.fr delphine.pflieger{at}cea.fr Abstract Full Text Info/History Metrics Preview PDF Abstract The response to targeted therapies and immune checkpoint inhibitors for patients suffering from metastatic clear cell renal cell carcinoma (ccRCC) is heterogeneous and currently not predictable in clinic. In this work, a comprehensive integrated study of 700 ccRCCs profiled by DNA methylation and RNA sequencing showed that the hyper-methylated tumors exhibited a worse prognosis, a higher fraction of cycling tumor cells and a lower activity of homeobox transcription factors. To translate the use of DNA methylation information into a clinical setting, we developed a simple model accurately predicting the ccRCC methylation subtypes (AUC-ROCs of 0.91) from two gene expression ratios (IGF2BP3/PCCA, TNNT1/TMEM88). In addition, these methylation subtypes were significantly associated with the therapeutic outcome of patients to anti-PD-1, mTOR inhibitor or tyrosine kinase inhibitor therapies. Overall, our framework for predicting the ccRCC DNA methylation subtypes from targeted gene expression data is easy to translate in clinic and contributes to better personalization of ccRCC therapies. Introduction Renal cell carcinoma (RCC) is one of the ten most diagnosed cancers worldwide. Three major RCC subtypes have been described with a prevalence greater than 5%. The most common RCC is clear cell RCC (ccRCC) occurring in 80% of cases, whereas papillary RCC (pRCC) and chromophobe RCC (chRCC) make up ∼20% of kidney cancers 1 . For localized ccRCC, the sole curative treatment remains partial or radical nephrectomy 1 , 2 while the advent of targeted therapies, such as tyrosine kinase inhibitors (TKI), mTOR inhibitors, and immunotherapies, has significantly improved survival rates for patients with metastatic disease 3 , 4 . However, the response to targeted and immuno-therapies are variable, with many patients not gaining a durable response and suffering unnecessary serious side effects 5 – 10 . There is currently no approved predictive molecular biomarker or score to guide selection of the most appropriate therapy; this is an area of unmet clinical need. The integration of comprehensive molecular profiling techniques, encompassing genomic, transcriptomic, proteomic and DNA methylation assays, have greatly enhanced our understanding of ccRCC tumor biology 11 – 15 . However, despite these advancements, the elucidation of the determinants of treatment outcome remains a complex and ongoing challenge. In order to select the most suitable treatment for cancer patients, the quantification of PD-L1 protein by immunohistochemistry (IHC) is employed in clinical practice as an FDA-approved companion biomarker for some cancers 16 . However, PD-L1 protein quantification does not definitively predict response to treatment and exhibits a significant proportion of inaccurate predictions 17 . Similarly, while Tumor Mutation Burden (TMB) is correlated with the response to immune checkpoint inhibitors in several cancers such as NSCLC or malignant melanoma, its predictive value is limited in kidney cancer which harbors a lower mutation rate 18 – 20 . Predictive scores based on gene expression data have been developed in several cancer types including JAVELIN 20 for ccRCC samples, or MIAS 21 , GEP 22 , MHC-I and MHC-II 23 and IMPRES 24 in melanoma. However, to our knowledge, none of these gene expression-based scores, requiring computational analyses, are used in clinical practice 25 , 26 . Alongside gene expression data, previous studies have established the clinical relevance of DNA methylation data for cancer prognosis. Several analyses revealed that ccRCC was divided into two or three DNA methylation subtypes exhibiting different patient survivals 12 , 27 , 28 . Some previous works have developed prognostic scores based on methylation data to predict patient survival in ccRCC, but failed in additional cohorts and did not predict the response to targeted therapies or immunotherapies 28 – 30 . In addition, previous works have showed that the DNA methylation of tumors was associated with response to immunotherapy in non-small cell lung cancer (NSCLC), malignant melanoma, sarcoma and ccRCC 31 – 33 . However, although that the recent work of Lu et al. 33 was performed on ccRCC samples of patients treated by immunotherapy or a tyrosine kinase inhibitor (TKI), the classification they have developed based on methyl-array or gene expression data failed to associate patien progression-free survival and predicted methylation subtypes in additional cohorts. Clinical trials in ccRCC or melanoma and biomarker discovery have largely focused on gene expression level 6 , 20 , 22 , 24 . However, DNA methylation subtypes have shown their relevance for prognosis stratification in ccRCC 12 , 27 , 28 . Consequently, this experimental limitation restricts the direct comprehensive evaluation of DNA methylation data to predict patient responses to anti-tumor therapies. One approach to overcoming this data shortage is to infer the DNA methylation subtype of each patient using available tissue expression data, a cost-effective and easy-to-translate in clinic strategy to improve patient stratification. In our work, we performed a comprehensive meta-analysis of DNA methylation levels in 727 tumor samples to better understand the determinants of treatment response in metastatic ccRCC. We first investigated existing clustering methods of ccRCC tumors based on DNA methylation data signatures to determine the most suitable one in a prognostic setting. We then relied on cohorts of tissues profiled both by transcriptomics and methyl-array to decipher the molecular and cellular discrepancies between ccRCC methylation subtypes in terms of tumor microenvironment (TME) composition, key biological processes and transcription factor activities. Subsequently, we developed simple classification models based on gene ratios to predict the ccRCC methylation subtypes from gene expression data. Finally, the application of these predictive markers to transcriptomics data of tumor tissues from patients enrolled in clinical drug trials demonstrated that the identified DNA methylation subtypes were associated with the treatment response in patients with ccRCC. Results Meta-analysis of ccRCC clustering from DNA methylation data Prior studies have investigated the DNA methylation subtypes of ccRCC samples. In these works, Arai et al. 27 , Sato et al. 12 , Evelönn et al. 28 and Lu et al. 33 used unsupervised clustering methods to group ccRCC tumor samples based on beta values of probes from methyl-array data. They demonstrated an association between hyper-methylated profiles and unfavorable prognosis. Interestingly, these methods selected different probes as signatures to cluster ccRCC samples into different DNA methylation subtypes. In this study, we evaluated the clustering approaches of the four aforementioned DNA methylation studies to identify the optimal sample subdivision. Methyl-array and gene expression data were obtained for 757 and 2072 samples, from four and six data sources, respectively. We merged the four distinct methyl-array datasets into a comprehensive unified dataset comprising a total of 727 samples. Of note, patient age, gender and clinical stages were equally distributed between the four datasets (Table S1). In the process of merging datasets, some probes used in the methylation signatures from Arai et al., Sato et al. and Evelönn et al. were not retained in the unified dataset because they were not shared between the Illumina 450K and EPIC methyl-arrays or had too many missing values (Fig. S1). To assess the potential signal loss due to non-conservation of all the signature probes, we compared the clustering of ccRCC samples based on the original signature subsets of probes available in our unified cohort. Of note, the Lu et al. clustering method was used with the total number of probes and was thus not part of this analysis. To reproduce the Arai et al. clustering method, we used beta values from the TCGA dataset of matched normal and tumor samples (27K Illumina array, N = 199). We performed hierarchical clustering to divide ccRCC samples into High and Low methylation level subtypes based on the difference between beta values from tumor and normal matched samples of 731 probes from the published signature 27 . We observed that the clusters were highly consistent with the ones obtained with the subset of 531 beta values (one value per CpG probe) from the tumor samples (Accuracy = 85.43%) (Table S2). To assess the Sato et al. 12 and Evelönn et al. 28 methods, we used the Sato et al. dataset which consisted of 106 samples whose beta values were obtained with a 450K Illumina array platform. Remarkably, the cluster memberships generated by the Sato et al. method using a reduced signature of 1006 probes out of 1672 were highly consistent with the cluster memberships obtained in the original study 7 (Accuracy = 93.4%) (Table S3). Similarly, the Evelönn et al. method yielded comparable results when applied to a subset of 148 CpG probes available in the unified dataset, instead of the original signature consisting of 174 CpG probes (Accuracy = 93.4%) (Table S4). To compare the four aforementioned methods, we performed each clustering analysis on the unified cohort and evaluated their prognostic implications in terms of overall survival (OS). The Arai et al., Sato et al., Evelönn et al. and Lu et al. methods stratified the unified cohort into two distinct groups with 267, 109, 274 and 242 samples, respectively, identified as hyper-methylated ( Fig. 1A , Table S5). The groups of hyper-methylated tumors predicted by each of the four methods were associated with significantly shorter OS (two-sided log-rank tests P<1.9e-4, P<1.6e-10 and P<1e-5, P<2.2e-9 for Arai et al., Sato et al., Evelönn et al. and Lu et al. methods, respectively) ( Fig. 1B-E ). In fact, the Sato et al. method, for which 15% of ccRCCs were hyper-methylated (109 out of 727 samples), yielded the most significant subdivision according to patient prognosis and was consequently selected for downstream analyses. Of note, the Lu et al. signature was OS-driven 33 , which was consistent with the fact that this method exhibited the second most significant OS difference. However, the Sato et al. groups were based on features correlated with tumor hyper-methylation in an unsupervised manner, and thus could be defined as molecular subtypes. Download figure Open in new tab Figure 1: Comparative survival analyses of ccRCC DNA methylation subtypes. A . Clustering of ccRCC samples from a unified cohort of four datasets into hyper- (high) and hypo-methylated (low) subtypes arranged according to the Sato et al. method. The 1,006 beta-values from the Sato et al. signature are visualized by the heatmap. Overall survival analysis between the methylation subtypes following the B . Arai et al., C . Sato et al., D . Evelönn et al. and E . Lu et al. methods. F . Overall survival analysis between hyper- and hypo-methylated tumors divided by the Sato et al. method in low-risk (stages I and II) samples. G . Overall survival analysis between hyper- and hypo-methylated tumors divided by the Sato et al. method in high-risk samples (stages III and IV). OS values are in months. Interestingly, we observed that the hyper- and hypo-methylated ccRCC samples inferred from the Sato et al. method were characterized by higher methylation levels compared to matched normal adjacent tissues in TCGA data (450k Illumina array) (Fig. S2). Consequently, our use of the term hypo-methylation here is made strictly within the framework of comparing ccRCC samples to each other and discriminating tumors that are not hyper-methylated. We observed that methylation subtypes in low-risk patients with clinical stages I or II did not yield significant differences in OS ( Fig. 1F ). Conversely, stage III-IV patients with hyper-methylated tumors exhibited a lower overall survival than patients of the same stages with hypo-methylated tumors ( Fig. 1G ). This showed that the prognosis of high-risk ccRCC tumors is significantly associated with DNA methylation subtypes. However, tumor stage and DNA methylation subtypes are both significant in a multivariate Cox regression analysis (P<1e-3) (Fig. S3). Collectively, these results support that the ccRCC harbors distinct DNA methylation subtypes associated with contrasted prognoses. In summary, the comparisons of the different clustering methods based on different subsets of DNA methylation probes demonstrated a higher relevance of the Sato et al. method which we decided to use as a reference for downstream analyses. Exploring ccRCC DNA methylation subtypes Following the clustering of the unified cohort of ccRCC samples based on the Sato et al. subdivision method, we explored in detail the clinical, cellular and molecular features that distinguish the two methylation subtypes. While we observed similar gender balance across the different cohorts (Table S6), we found that the hyper-methylated subtype exhibited a lower proportion of tumors from women compared to the hypo-methylated subtype, with 16% and 28% of samples, respectively (Pearson’s chi-squared P=0.0035). Also, we explored the clustering of ccRCC samples based on beta values into three groups. However, this larger number of clusters did not add value to the prognostic profiling (Fig. S4). Then, to investigate the differences in transcriptomic profile and cellular composition of the tumor microenvironment (TME) between the two methylation subtypes, we retained TCGA, CPTAC and Sato et al. samples from our unified cohort with matched DNA methyl-array and gene expression data (N = 589). To study transcriptomic changes, we first carried out an unsupervised consensus clustering of k-means from gene expression data of each cohort separately and split ccRCC samples into the two methylation groups defined by Sato et al. We observed that gene expression profiles were barely correlated with methylation subtypes (accuracy = 63.16%) ( Fig. 2A “RNA clusters”, Table S7). Download figure Open in new tab Figure 2: Transcriptomic characterization of ccRCC DNA methylation subtypes. A . Based on the hyper-(high) and hypo-methylated (low) subtypes arranged according to the Sato et al. method, the heatmap visualizes the predicted cell fractions by CIBERSORTx and ssGSEA values of the biological pathway hallmarks (MSigDB). ‘RNA clusters’ refers to groups obtained by consensus clustering of gene expression values. B . Cell fraction values significantly different between hyper- (high) and hypo-methylated (low) subtypes according to Wilcoxon rank-sum tests. C . Transcription factor activity inferred using the VIPER algorithm based on the DoRothEA database of TFs and gene targets. A positive NES indicated higher activity while a negative NES indicated lower activity in hyper-methylated tumors. Here, the top 15 of TFs that yielded a significant difference between the methylation subtypes were retained for visualization purposes. D . Gene expression values of DNA methylatransferases between Hyper-(High) and Hypo-methylated (Low) subtypes E . Gene expression values of DNA demethylation enzymes between Hyper-(High) and Hypo-methylated (Low) subtypes. P -values in boxplots in B, D and E were calculated by Wilcoxon rank-sum tests and corrected by the Benjamini-Hochberg method. In B, P-values from the three datasets were pooled by the Stouffer method. Since TME composition has been described in previous works to strongly influence cancer prognosis and treatment response outcomes 34 – 36 , we investigated whether differences existed between ccRCC methylation subtypes. We used CIBERSORTx to perform cellular deconvolution based on transcriptomic data and ccRCC single-cell reference data refined in our previous work 37 . We predicted the fractions of 21 different cell types for each ccRCC sample (Tables S8-10) and observed that the TME of hyper-methylated samples was significantly enriched in cycling tumor cells and depleted in four cell types: tumor cells, CA9-high tumor cells, stromal cells, and endothelial tip cells ( Fig. 2A and B ). Interestingly, the fractions of 15 immune cells did not differ significantly between methylation subtypes (Tables S11-13). To further study the immunity content, we calculated a transcriptomic score per ccRCC sample reflecting the tertiary lymphoid structures (TLS) within the TME 34 and did not find a significant difference between the methylation subtypes (two-sided Wilcoxon rank-sum test P > 0.05) (Fig. S5, Table S14). These results supported the fact that the two ccRCC methylation statuses harbored similar TME immune compositions. To reveal molecular contrasts between ccRCC methylation subtypes in terms of biological pathways, we performed single-sample gene set enrichment analysis (ssGSEA) using the 50 hallmark and 189 oncogenic signature gene sets from the MSigDB database 38 , 39 . We identified 16 hallmark gene sets significantly deregulated between the ccRCC methylation clusters across the three datasets. Remarkably, among the top 5 most deregulated biological processes, we observed the unfolded protein response (corrected Stouffer P < 1e-21) and several cell cycle progression-related gene sets to be up-regulated in hyper-methylated tumors, namely the mTORC1 complex, designating the association of mTOR protein with several regulatory proteins such as raptor, mLST8, deptor and PRAS40 (corrected Stouffer P < 1e-21), the MYC oncogene (corrected Stouffer P < 1e-22), the G2M DNA damage checkpoint (corrected Stouffer P < 1e-21) and the E2F family of transcription factors (corrected Stouffer P < 1e-21) ( Fig. 2A , Table S15). Then, using oncogenic signature gene sets, we observed 30 significant pathways across the three datasets that showed a high discrepancy in oncogenic processes between the methylation subtypes (Fig. S6, Tables S16). Among the top 5 most significant gene sets, we found in the hyper-methylation subtype an up-regulation of E2F1 transcription factor (E2F1_UP.V1_UP, corrected Stouffer P <1e-18) and a down-regulation of GLI1 which is specific of kidney epithelium cells and mediates Hedgehog signaling (GLI1_UP.V1_UP, corrected Stouffer P < 1e-19). Interestingly, genes up-regulated by the mTOR inhibitor everolimus (MTOR_UP.V1_UP) were up-regulated in hyper-methylated tumors (corrected Stouffer P < 1e-17), while genes annotated as down-regulated by this drug were found down-regulated in this sub-group (MTOR_UP.N4.V1_DN, corrected Stouffer P < 1e-17) (Fig. S6). To further investigate the transcriptomic profiling of the two ccRCC methylation subtypes, we used the VIPER algorithm and the Dorothea database to calculate the transcription factor (TF) activity of 271 TFs 40 , 41 . A TF was considered to be over-activated if the genes it promotes were over-expressed and the genes it inhibits were under-expressed. The p-values and normalized enrichment scores (NES) calculated for the three cohorts were combined by the Stouffer’s method and by averaging, respectively. We detected 3 over-activated (NES > 0) and 19 under-activated (NES < 0) TFs ( Fig. 2C , Table S17). Among the top 20 significant TFs, cell cycle-related E2F4 was over-activated (corrected Stouffer P < 1e-16) whereas the homeobox proteins that promote stem cell pluripotency and tumorigenesis NANOG, PBX Homeobox 3 (PBX3) or POU Class 5 Homeobox 1 (POU5F1) 42 – 44 were among the TFs that were most under-activated in hyper-methylated tumors (corrected Stouffer P < 1e-10, P < 1e-10 and P < 1e-8, respectively). Interestingly, the NANOG gene expression was not correlated with its decreased activity in the hyper-methylated group in the TCGA (two-sided Wilcoxon rank-sum test P=0.45) and CPTAC (two-sided Wilcoxon rank-sum test P = 0.12) datasets while the expression values were significantly different in the Sato et al. dataset (two-sided Wilcoxon rank-sum test P = 6.9e-3) (Fig. S7). In addition, the activity of CREB3L1 was higher in the hyper-methylation subtypes, which was consistent with the up-regulation of the unfolded protein response ( Fig. 2A, C ), known to be directly regulated by CREB3L1 45 . Conversely, we observed a lower activity of GATA TFs ( GATA2, GATA3, GATA6 ) which are involved in increased tumor cell invasiveness and associated with poor outcome in ccRCC 46 – 49 . The deregulation of such processes and transcription factors suggested that the hyper-methylated tumors were characterized by the deregulation of NANOG and associated proteins. Additionally, the over-activation of unfolded protein response in these hyper-methylated tumors may indicate an increase of the levels of misfolded proteins in the endoplasmic reticulum. Cytosine methylation of DNA is regulated by the concerted action of DNA methyltransferases (DNMTs) and demethylases. We assessed the gene expression of these enzymes between the ccRCC DNA methylation subtypes. Among the three DNMTs that are responsible for de novo or maintenance DNA methylation, only DNMT3B , involved in de novo methylation, was overexpressed across the three datasets (two-sided Wilcoxon rank-sum tests P<0.001) ( Fig. 2D ). DNMT3A and DNMT3L involved in de novo methylation and DNMT1 in maintenance methylation were not deregulated between methylation subtypes. Conversely, among the three DNA demethylase (ten-eleven-translocation proteins, TET1, TET2, TET3 ), we observed a significantly lower gene expression of TET2 in hyper-methylated tumors across datasets (two-sided Wilcoxon rank-sum tests P<0.01) ( Fig. 2E ) while TET1 and TET3 expression values did not yield significant differences between subtypes. These results suggested that the DNA hyper-methylation in ccRCC may be due to both an increase of the DNA methyltransferase DNMT3B and a decrease of the DNA demethylase TET2 gene expression levels. Of note, protein abundances of DNMT3B and TET2 were quantified for a too small number of CPTAC samples (16 samples) and thus their difference between ccRCC methylation subtypes could not be assessed. Overall, these results demonstrated a high transcriptomic and cellular heterogeneity between the ccRCC methylation subtypes. An increased activity of cell cycle pathways and a higher fraction of cycling tumor cells may be involved in the difference of survival observed between these subtypes. Moreover, two enzymes involved in DNA methylation ( DNMT3B, TET2 ) were deregulated and may be responsible for the hyper-methylation of ccRCC. Prediction of ccRCC methylation subtypes from gene expression data Having previously observed that patients with advanced ccRCCs (stages III-IV) harbored a lower overall survival when exhibiting hyper-methylated tumors ( Fig. 1G ), we sought to investigate whether ccRCC methylation status could influence patient response to systemic therapies. As molecular profiling of DNA methylation is not routinely performed in drug clinical trials, we first developed a cross-omics bioinformatics approach to predict the methylation subtype of ccRCC samples from gene expression data. We used the Sato et al. micro-array and the TCGA RNA-Seq datasets of 101 and 301 samples, respectively, along with the methylation subtype memberships of all samples. To select the most discriminant genes between the two methylation subtypes, we first performed two-sided Wilcoxon rank-sum tests between hyper- and hypo-methylated groups of the two cohorts. Then, we ranked genes according to p-values and selected overlapping differentially expressed genes (DEGs) between the top 150 DEGs of each cohort (Tables S18, S19). In that way, a subset of 24 DEGs between high and low methylation subtypes was retained. For each gene expression feature, we determined the best cutoff value dividing samples into high or low methylation subtypes using the Youden’s index method. Then, we calculated areas under the receiver operating characteristic curve (AUC-ROC), Balanced accuracy, and Matthews correlation coefficient (MCC) as general performance metrics ( Fig. 3 , Tables S20, S21). Download figure Open in new tab Figure 3: General workflow to predict ccRCC DNA methylation subtypes from gene expression data. A . The 24 genes that overlapped the top 150 DEGs between the methylation subtypes obtained by Wilcoxon rank-sum tests into the Sato et al. and the TCGA datasets were retained. Then, the Youden’s index method was used for each gene (or gene ratio) to determine the best value threshold to divide the samples between High or Low methylation profiles. B . The same features and thresholds were applied to the CPTAC data as a validation set survival analysis was performed between the predicted methylation subtypes to compare with the ground truth based on methyl-array data. C . The top 2 gene expression ratios on three datasets of patients included into clinical drug trials were selected. Due to a high difference of gene ratio value distributions across the datasets, we used a custom threshold which was the expected proportion of hyper-methylated samples (15%) according to our results from our meta-analysis of methyl-array data. Finally, the treatment response between the predicted DNA methylation subtypes was compared using survival analysis. *Gene ratios were calculated by combining the 24 DEGs two-by-two. We found that the top 10 single genes harbored strong performance in predicting high and low methylation subtypes in the training set (TCGA dataset), as evidenced by high AUC-ROC (from 0.81 to 0.91), MCC (from 0.40 to 0.58) and balanced accuracy (from 0.75 to 0.85) values ( Fig. 4A ). To ensure the reliability of these predictors, we similarly tested these genes on an independent validation cohort consisting of 187 samples from the CPTAC dataset using the same gene expression thresholds determined from the TCGA training set. We observed that these genes still demonstrated high predictive values, with AUC-ROC values ranging from 0.80 to 0.91 ( Fig. 4B ). As expected, the TCGA expression thresholds applied on CPTAC dataset resulted in slightly lower classification performances compared to the training perspective: the MCC and balanced accuracy values ranged from 0.31 to 0.58 and from 0.59 to 0.85, respectively. To estimate the prognostic value of these genes, we selected the two genes with the best MCC values in the TGCA training set, which were IGF2BP3 and TNNT1 , and evaluated their prognostic power on the CPTAC validation cohort. We observed that methylation subtypes predicted by these two genes significantly separated patients from the CPTAC cohort based on their overall survival, with a hyper-methylation group associated with poor survival (two-sided log-rank test P < 0.05) ( Fig. 4C-E ), which was consistent with the methylation-based analysis. Therefore, after a two-step training-validation process, IGF2BP3 and TNNT1 emerged as robust markers of ccRCC methylation subtypes, associated with patient prognosis. Download figure Open in new tab Figure 4: Assessment of single gene expression values to classify ccRCC samples into DNA methylation subtypes. Only the top 10 DEGs were visualized. Classification performances on the A . TCGA (“training” for the thresholds) B . CPTAC (validation) datasets. Survival analysis of DNA methylation subtypes obtained by C . the Sato et al. method applied on methyl-array data (ground truth), D . IGF2BP3 and E . TNNT1 gene expression values. AUC: areas under the receiver operating characteristic curve MCC: Matthews correlation coefficient. BalancedAcc: balanced accuracy. PPV: positive predictive value. NPV: negative predictive value. OS values are in months. We next tried to investigate a possible link between promoter methylation level and expression for the 24 DEGs identified between high and low methylation subtypes. We found promoter-associated DNA methylation probes for 5 genes ( IGF2BP3, RNF152, TMEM88, AURKB, HJURP ). We observed that the averaged methylation levels of the promoter-associated probes related to IGF2BP3, TMEM88 and AURKB yielded significant differences between hypo-and hyper-methylated tumors (two-sided Wilcoxon rank-sum test P < 1e-14, P = 0.04 and P < 1e-2, respectively) (Fig. S8A). Next, we summarized the methylation levels of the promoters of these genes by averaging the beta values of the corresponding probes. We found that lower methylation of the IGF2BP3 promoter in hyper-methylated tumors was associated with an increase in its expression in the TCGA (Spearman correlation test rho=-0.46, P < 2.2e-16), CPTAC (Spearman correlation test rho=-0.43, P < 1e-9) and Sato et al. (Spearman correlation test rho=-0.49, P < 1e-6) datasets (Fig. S8 B-D). Interestingly, we observed a decreased methylation level of IGF2BP3 promoter-associated probes in hyper-methylated samples compared to normal adjacent tissue (NAT), whereas the level of methylation did not yield a significant difference between hypo-methylated tumors and matching NAT (Fig. S9). Similarly, we observed that a higher methylation of the TMEM88 promoter in hyper-methylated tumors (two-sided Wilcoxon rank-sum test P<0.05) was linked to a decrease of its expression across the TCGA (Spearman correlation test rho=-0.34, P < 2.2e-10), CPTAC (Spearman correlation test rho=-0.39, P < 1e-8) and Sato et al. (Spearman correlation test rho=-0.43, P < 9.3e-6) cohorts (Fig. S8 E-G). However, we did not observe a significant correlation between the methylation level of the AURKB promoter and its expression (Fig. S8 H-J). In conclusion, we could only retrieve the promoter methylation level of 5 genes among the 24 retained DEGs between high and low methylation subtypes. Among them, we observed an association between the increased/decreased promoter methylation and decreased/increased expression of the two genes TMEM88 and IGF2BP3 , respectively. Prior studies revealed that gene ratios constitute robust and accurate predictors of biological subtypes 50 – 52 . Thus, we investigated their use for predicting ccRCC methylation sub-groups. First, we calculated ratios for each pair of the 24 DEGs between methylation subtypes (Tables S22, S23). We next evaluated the classification performance of each gene ratio in our TCGA training dataset. The best threshold values to divide the cohort between hyper- or hypo-methylated subtypes were determined by the Youden’s index method. We observed that the top 10 gene ratios yielded high predictive performances in the training dataset with AUC-ROCs ranging from 0.89 to 0.94, MCC ranging from 0.61 to 0.65 and balanced accuracy ranging from 0.85 to 0.88 ( Fig. 5A , Table S24). Furthermore, to ensure the robustness of the gene ratios, the thresholds of gene expression optimized from the TCGA training dataset were directly applied to the CPTAC validation dataset. We observed that the top 10 gene ratios conserved highly predictive performances with AUC-ROCs ranging from 0.90 to 0.94, MCC ranging from 0.48 to 0.62 and a balanced accuracy ranging from 0.78 to 0.83 ( Fig. 5B , Table S25). We also estimated the prognostic values of these gene rations by selecting the top 2 gene ratios based on MCC values, IGF2BP3/PCCA and TNNT1 / TMEM88 , and evaluated their ability to cluster patients in the CPTAC validation cohort based on prognosis. Interestingly, we found that similarly to the ground truth of the clusters obtained from methyl-array values ( Fig. 5C ) the p-values obtained for survival analyses between methylation subtypes predicted by IGF2BP3 / PCCA (two-sided log-rang test P < 1e-4) and TNNT1 / TMEM88 (two-sided log-rang test P = 0.0019) ( Fig. 5D-E ) were more significant than the ones calculated using single-gene expression of IGF2BP3 and TNNT1 ( Fig. 4D-E ). This illustrated the relevance of gene ratios for predicting ccRCC methylation subtypes and their association with patient prognosis. Download figure Open in new tab Figure 5: Assessment of gene expression ratios to classify ccRCC samples into DNA methylation subtypes. Only the top 10 gene ratios were considered. Classification performances on the A . TCGA (“training” for the thresholds) B . CPTAC (validation) datasets. Survival analysis of methylation subtypes obtained by C . the Sato et al. Method applied on methyl-array data (ground truth), D . IGF2BP3/PCCA and E . TNNT1/TMEM88 ratios. AUC: areas under the receiver operating characteristic curve MCC: Matthews correlation coefficient. BalancedAcc: balanced accuracy. PPV: positive predictive value. NPV: negative predictive value. OS values are in months. To assess the effect of intratumoral heterogeneity (ITH) on subtypes prediction, we selected 40 patients from the CPTAC dataset with at least two biopsies from the primary tumor site for a total of 131 samples. We observed that predictions of DNA methylation subtypes by a gene ratio ( IGF2BP3/PCCA and TNNT1 / TMEM88 ) on one sample are consistent across samples for a given patient with only 3 and 4 patients with discrepancy out of 40 for IGF2BP3/PCCA and TNNT1/TMEM88 , respectively (Fig. S10, Table S26). Moreover, we investigated the localization of these four genes IGF2BP3/PCCA IGF2BP3, PCCA, TNNT1 and TMEM88 in the single-cell dataset used in previous analyses ( Fig. 2B ). TNNT1 was not expressed in this data while IGF2BP3 could not be associated with a specific cell type due to weak expression values. However, we found that PCCA was especially expressed in the tumor cells CA9 -High while TMEM88 was strongly associated to endothelial and tip cells, which gave some insight into the TME-level information contained in these gene ratios (Fig. S11). Overall, we demonstrated that gene expression values and gene ratios effectively predicted DNA methylation subtypes in ccRCC, paving the way for exploring the methylation status in clinical drug trials of patients profiled only by tumor transcriptome. Association of ccRCC methylation subtypes with response to targeted therapies and immunotherapy We investigated the ability of the gene expression ratios IGF2BP3 / PCCA and TNNT1 / TMEM88 to predict the response of ccRCC patients to targeted and immuno-therapies. We used publicly available gene expression data from the CheckMate cohort 9 comprising 309 primary or metastatic site biopsies of ccRCC from patients treated by nivolumab (anti-PD-1) or everolimus (mTOR inhibitor) and from patients treated with sunitinib (TKI) from Beuselinck et al. 53 , Zhang et al. 15 , Motzer et al. 8 and Banchereau et al. 54 with 53, 51, 823 and 247 ccRCC samples, respectively. Moreover, to comprehensively characterize these data, we divided the CheckMate cohort into four subgroups according to treatment (anti-PD-1 nivolumab or mTOR inhibitor everolimus) and tissue type (primary or metastatic site samples). To classify patients based on gene ratio values, we could not use previously estimated thresholds because of a high discrepancy between TCGA data and the gene expression distributions from these independent datasets (Fig. S12). These differences could be attributed to the data processing and batch effect corrections performed in the previous work 9 or by the difference of sequencing technology 53 . Therefore, we determined the threshold to divide each cohort into two methylation subtypes using the proportion of 15% of hyper-methylated profiles observed in our analysis ( Fig. 1A ). The best gene ratio for a subgroup was retained based on the most significant log-rank test p-value calculated by comparing progression-free survival (PFS) values between predicted methylation subtypes. From the CheckMate cohort, we found that the primary site tumors predicted to be hyper-methylated using gene ratios (Table S27) were enriched in good responders to nivolumab ( Fig. 6A , two-sided Fisher’s exact test P < 0.05) with significant differences in PFS (two-sided log-rank test P < 0.05) ( Fig. 6F ). Download figure Open in new tab Figure 6: Association of ccRCC DNA methylation subtypes with response to immunotherapy and mTOR inhibitor. A . Comparison of the proportions of patients associated with clinical benefit (CB), intermediate clinical benefit (ICB) and non-clinical benefit (NCB) for treatments with everolimus or nivolumab according to the DNA methylation subtypes predicted by gene ratio. B . Proportions of cancer recurrence events between predicted DNA methylation subtypes by treatment. Progression-Free survival (PFS) analysis between predicted ccRCC methylation subtypes of C . metastatic samples treated with everolimus, D . primary samples treated with everolimus, E . metastatic samples treated with nivolumab, F. primary samples treated with nivolumab. PFS values are in months. Similarly, we observed that the hyper-methylated group treated by nivolumab was enriched in recurrence-free samples (26.6%) compared to the hypo-methylated group (4.8%) (two-sided Fisher’s exact test P < 0.01) ( Fig. 6B ). Interestingly, primary site tumors predicted to be hyper-methylated were also enriched in poor responders to everolimus ( Fig. 6A ) with significant differences in PFS (two-sided log-rank test P < 0.01) ( Fig. 6D ). However, we did not observe significant differences in terms of OS between the methylation groups obtained from the primary site samples (Fig. S13A, C). Regarding tumor metastases, the hyper-methylation group predicted by the gene ratios was enriched in poor responders to everolimus (Pearson’s two-sided Fisher’s exact test P < 0.05) with significant differences in PFS (two-sided log-rank test P < 0.05) ( Fig. 6A, C ) and also tended to be associated with a poor response to nivolumab although non significantly (PFS two-sided long-rank test P = 0.084) ( Fig. 6E ). Conversely, patients treated with nivolumab harbored significant differences only in terms of OS when metastatic samples were employed to determine the methylation subtypes (Fig. S13D). These observations suggested that the hyper-methylated primary and metastatic samples was a marker of poor response to everolimus. In contrast, hyper-methylated primary samples were associated with a good response to nivolumab. From the primary site tumors of patients treated with sunitinib in the Beuselinck et al. dataset, we observed that hyper-methylated tumors predicted by TNNT1 / TMEM88 ratio (Table S28) were enriched in progressive disease (PD) patients (two-sided Fisher’s exact test P < 0.05) ( Fig. 7A ) and exhibited shorter PFS (two-sided log-rank test < 0.01) ( Fig. 7B ). We obtained the same enrichment in PD patients from hyper-methylated tumors for the cohort of patients treated with sunitinib from the Zhang et al. dataset ( Fig. 7C , Table S29), although the proportion test was not significant. Also in this dataset we found that hyper-methylated tumors were significantly associated with lower PFS (two-sided log-rank test < 0.05) ( Fig. 7D ). In addition, we observed the same patterns in the Motzer et al. cohort of 319 patients treated with sunitinib with a higher proportion of poor RECIST responders and a lower PFS in the hyper-methylated subtype (two-sided Fisher’s exact test P < 0.05, two-sided log-rank test < 0.0001) ( Fig. 7E, F , Table S30). The same results were obtained in the Banchereau et al. dataset of 247 ccRCC patients treated with sunitinib for the RECIST (two-sided Fisher’s exact test P < 0.05) and PFS (two-sided log-rank test = 6e-04) information ( Fig. 7G, H , Table S31). Download figure Open in new tab Figure 7: Association of ccRCC DNA methylation subtypes with response to sunitinib. A . Comparison of the proportions of treatment responses (RECIST) from the Beuselinck et al. cohort followed after sunitinib treatment, according to DNA methylation subtypes predicted by the TNNT1/TMEM88 expression ratio. B . PFS analysis of the methylation subtypes predicted by the TNNT1/TMEM88 ratio for the Beuselinck et al. cohort. C . Comparison of the proportions of treatment responses (RECIST) from the Zhang et al. cohort followed after sunitinib treatment in each DNA methylation subtype predicted by the TNNT1/TMEM88 ratio. D . PFS analysis of the methylation subtypes predicted by the TNNT1/TMEM88 ratio for the the Zhang et al. cohort. E . Comparison of the proportions of treatment responses (RECIST) from the Motzer et al. cohort followed after sunitinib treatment in each DNA methylation subtype predicted by the TNNT1/TMEM88 ratio. F . PFS analysis of the methylation subtypes predicted by the TNNT1/TMEM88 ratio for the the Motzer et al. cohort. G . Comparison of the proportions of treatment responses (RECIST) from the Banchereau et al. cohort followed after sunitinib treatment in each DNA methylation subtype predicted by the TNNT1/TMEM88 ratio. H . PFS analysis of the methylation subtypes predicted by the TNNT1/TMEM88 ratio for the the Banchereau et al. cohort. I . Proportions of cancer recurrence events between predicted DNA methylation subtypes for all sunitinib-treated patients. J . PFS analysis of methylation subtypes for patients with an intermediate MSKCC score in the Motzer et al. dataset. K . PFS analysis of methylation subtypes for PD-L1-positive in the Motzer et al. dataset. RECIST categories: complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD), inevaluable (NE). PFS values are in months. Finally, by combining the recurrence information across the Beuselinck et al., Zhang et al., Motzer et al. and the Banchereau et al. cohorts of sunitinib-treated patients, we found that hypo-methylated tumors were significantly enriched in recurrence-free patients (26% and 7% for the hypo- and hyper-methylated subtypes, respectively) (two-sided Fisher’s exact test P = 0.001) ( Fig. 7G ). The MSKCC prognostic model is widely used in the clinic to subdivide patients into low, intermediate or high risk 55 . The intermediate risk group is clinically particularly challenging, and we therefore assessed if defining the methylation subtypes may provide added value in this cohort of patients. In fact, the intermediate group treated with sunitinib (which comprises a majority of patients) is strongly divided into two PFS groups in the Motzer et al. dataset (two-sided log-rank test < 0.001) ( Fig. 7J , Fig. S14). In contrast, we observed no association between methylation subtypes and the combined treatment with atezolizumab+bevacizumab (anti-PD-L1+anti–VEGF) in the Motzer et al. dataset, neither in primary (Fig. S15, S16) nor in metastatic samples (Fig. S17). The same analyses in the Banchereau et al. dataset of primary samples treated with atezolizumab+bevacizumab or atezolizumab alone yielded the same results (Fig. S18). Thus, these results indicated that the predictive power of the DNA methylation subtypes is related to specific treatment modalities and not merely a consequence of their prognostic power. Finally, the quantification of PD-L1 protein is used as a companion biomarker of immunotherapy response. However, it may reflect immunity-related features of TME involved in larger response mechanisms. Therefore, we evaluated the two groups based on PD-L1 protein quantification (PD-L1-positive and PD-L1-negative) and methylation subtypes predicted by the TNNT1 / TMEM88 ratio. Interestingly, we observed that the groups of sunitinib-treated patients that are PD-L1-positive could be significantly divided by the TNNT1 / TMEM88 ratio with a worse PFS value for the hyper-methylated tumors contrary to PD-L1-negative patients in the Motzer et al. dataset (two-sided log-rank test = 1e-04) ( Fig. 7K , Fig. S19) and in the Banchereau et al. dataset (two-sided log-rank test = 0.0026) ( Fig. 7K , Fig. S20). To conclude, we showed that DNA methylation subtypes predicted from gene expression ratios, especially TNNT1 / TMEM88 , were associated with the patient response to mTOR inhibitor, TKI and anti-PD-1 therapies in ccRCC. More precisely, while predicted hyper-methylated tumor samples from the primary and metastatic sites had a poorer response to everolimus and sunitinib, primary samples predicted to be hyper-methylated harbored a good response to nivolumab. The prediction of DNA methylation subtypes from gene expression ratios is therefore an effective approach to add value to clinical features used currently in clinic like the MSKCC score and PD-L1 protein quantification, paving the way towards a better personalization of treatment in a clinical setting. Discussion In this paper, we congregated DNA methylation data from four studies resulting in a cohort of 727 patients with ccRCC. When applying on this merged cohort the different unsupervised clustering methods used in former studies, we concluded that the subdivision method into hyper- and hypo-methylated tumor tissues from Sato et al. 12 , based on 1,006 CpGs, was superior to other methylation clustering methods in prognosticating patients, particularly with stages III-IV ccRCC. Having selected the best-performing clustering method, we carried out an in-depth characterization of the biological differences between these subtypes. We found that ccRCC methylation subtypes differed in their expression of key regulators of global methylation levels, namely DNA methyltransferases (DNMTs) and demethylases (TETs). In particular, DNMT3B and TET2 were significantly over- and under-expressed, respectively, in hyper-methylated tumors. These observations suggest that the methylation differences observed between the subtypes may be caused by deregulation of DNMT3B and TET2 which are involved in de novo DNA methylation and demethylation processes, respectively, and may be involved in tumor development 56 , 57 . Having categorized the tumors into hypo- and hyper-methylated subgroups, we analyzed the proportions of cell types in the TME by using cellular deconvolution and we performed an analysis of biological pathways and of transcription factor activity for 589 samples with matched methylation and gene expression data. Based on these analyses, we aimed to assess the major biological shift between the hypo- and hyper-methylated subtypes. The TME of hyper-methylated tumors was composed of a higher fraction of cycling tumor cells (PCNA and MKI67-positive) characterized by a high cell growth/division rate 37 , 58 . By contrast, the hypo-methylated tumors harbored higher fractions of endothelial tip cells involved in angiogenesis, which is consistent with the recent investigation of methylation subtypes in ccRCC 33 . In addition, the gene expression analyses highlighted that the hyper-methylated tumor profile was deregulated in proliferation and transcription regulators, with a higher activity of MYC, mTOR, G2M checkpoint, E2F1 and PKCA and a lower activity of GLI1. Similarly, the oncogene EIF4GI involved in translation initiation as well as the E2F4 transcription factor were over-activated in the hyper-methylated subtype. Interestingly, the hyper-methylated samples exhibited a decreased activity of TCF4 (Immunoglobulin Transcription Factor 2) and of homeobox genes NANOG, PBX3 and POU5F1 related to tumorigenesis 42 – 44 . Notably, the down-regulation of homeobox genes observed in hyper-methylated tumors is consistent with the work from Sato et al. where the homeobox gene family was found to be over-represented in the genes that contain the signature probes 12 . Moreover, a recent work employed the expression values of homeobox genes as prognostic markers of ccRCC patients and showed that the down-regulation of several homeobox genes, in particular HOXD8, was correlated with tumor progression and poor prognosis 59 . Besides, the unfolded protein response was over-activated in the hyper-methylated tumors, which may be involved in tumour progression and limit the effectiveness of treatments 60 , 61 . We next sought to understand whether ccRCC methylation subtypes could shed light on the drug response of patients with advanced ccRCC. However, clinical drug trials including patients with metastatic ccRCC are devoid of methylation data, yet are instead associated with genomic and transcriptomic profiling of tumor tissues. Therefore, we developed a cross-omics classification method to predict ccRCC DNA methylation subtypes from gene expression data only. Remarkably, single gene expression (IGF2BP3, TNNT1) and gene ratios (IGF2BP3/PCCA, TNNT1/TMEM88) showed high capability to predict methylation subtypes on an independent validation dataset. We then used these gene expression ratios to predict the DNA methylation subtypes on three independent cohorts of ccRCC samples from patients treated with targeted therapies, such as mTOR inhibitor (everolimus), anti-VEGF (sunitinib, bevacizumab) and anti-PD-1 immunotherapy (nivolumab, atezolizumab). The hyper-methylated subtype predicted by IGF2BP3/PCCA or TNNT1/TMEM88 ratios was associated with worse outcomes of everolimus and sunitinib treatment from both primary site tumors and metastases. One may speculate that the lack of clinical benefit with the sunitinib in the hyper-methylated group was associated with our observation that these tumors were characterized by lower fractions of endothelial tip cells involved in neoangiogenesis, presumed targets of the anti-angiogenic based therapies, leading to an increased resistance to anti-VEGF treatments which target pro-angiogenic receptors. Regarding the response to nivolumab, while the hyper-methylation status of metastases tended to be associated with a worse outcome, it was on the other hand significantly associated with a favorable patient response when the primary tumor sites was considered. Of note, these contrasts in clinical benefits between treatments according to the methylation subtypes and the fact that these subtypes yielded no significant results for other treatments (atezolizumab+bevacizumab or atezolizumab) ensure the fact that the gene ratios were specifically related to treatment response and not only prognosis. Besides, these results on nivolumab, everolimus and atezolizumab treatments should be confirmed in additional cohorts to confirm the relevance of predicted methylation subtypes in such settings. Interestingly, the Insulin-Like Growth Factor 2 mRNA Binding Protein 3 (IGF2BP3) exhibited the most discriminating gene expression pattern to classify ccRCC tumors between hyper- and hypo-methylated subtypes with a higher expression in the hyper-methylated tumors. Remarkably, we also observed that the gene expression of IG2BP3 was negatively correlated with its promoter methylation, in agreement with recent works showing that demethylation of IG2BP3 promoters induced higher gene expression 62 which is involved in tumorigenic activity 62 – 64 . Moreover, we demonstrated in our work that DNA methylation of the IGF2BP3 promoter in hyper-methylated ccRCC tumors was lower than in matching normal samples, whereas it was similar between hypo-methylated tumors and matching NAT samples. IGF2BP3 is an RNA-binding protein which targets around 1000 to 4000 transcripts, is largely described to be a tumor-promoting factor and is over-expressed in many cancers; it leads to drug resistance 65 , 66 and to a poor prognosis 67 , 68 and it induces epithelial-mesenchymal transition in esophageal squamous cell carcinoma cells 69 . Also, the Transmembrane Protein 88 ( TMEM88 ) which was part of the TNNT1/TMEM88 ratio associated especially with everolimus and sunitinib treatments with a poorer response for higher values. TMEM88 is involved in the inhibition of the canonical Wnt pathway and may play a role in tumor initiation and progression 70 . In our work, we showed that its gene expression was especially localized in endothelial cells. This again underlines the fact that angiogenesis is an important component of the methylation subtype discrepancy. In conclusion, these findings suggested that the hyper-methylation of ccRCC, observed prior to drug therapy, was correlated with a reduced angiogenesis, a decreased regulatory activity of homeobox genes, an increase in tumor proliferation and a heightened IG2BP3 activity. These biological features were associated with different patient survival and drug response. Moreover, our work highlighted that molecular discrepancies between DNA methylation subtypes of ccRCC tumors predicted by simple gene expression ratios (IGF2BP3/PCCA or TNNT1/TMEM88) were strongly associated with prognosis and response to targeted therapies and anti-PD-1 immunotherapy. In addition, the simplicity of gene ratio quantification (e.g., by qRT-PCR) could lead to an easy translation in a clinical setting. These results contribute to a better understanding of the treatment response determinants in ccRCC patients, and ultimately may lead to better personalization of therapies in clinical practice. Materials and Methods Methyl-array data and data filtering Methyl-array data of ccRCC samples from 4 independent patient cohorts were collected (Fig. S1), comprising data from The Cancer Genome Atlas (TCGA) (27K Illumina array, N=199 or 450k Illumina array, N=318) 11 , Clinical Proteomic Tumor Analysis Consortium (CPTAC) (methyl-array EPIC Illumina array, N=196) 13 , 14 , Sato et al. dataset (450k Illumina array, N=111) 12 and Evelönn et al. dataset (450k Illumina array, N=132) 28 . In the case of tumor samples from matched patients in the TCGA datasets, only one was retained with priority to the 450k Illumina array for matching samples across platforms. The beta values with more than 15% of missing values and then samples with more than 20% of missing values were removed. Remaining missing values in methyl-array datasets were imputed by the knn() function from the VIM R package (v. 6.2.2). The batch effect between the methyl-array datasets was corrected by the ComBat algorithm from the R/Bioconductor package sva (v. 3.42.0). Gene expression data Gene expression data of 2072 samples from five data sources were used in our study: The Cancer Genome Atlas (TCGA) (N = 301, RNA-Seq, in transcripts per million) 6 , Clinical Proteomic Tumor Analysis Consortium (CPTAC) (N = 187, RNA-Seq, in transcripts per million) 8 , 9 , the Sato et al. dataset (N=101, micro-array) 12 , the Braun et al. dataset (N = 309, RNA-Seq, normalized transcripts per million) 9 , the Beuselinck et al. dataset (N=53, micro-array) 53 , the Zhang et al. dataset (N=51) 15 , the Motzer et al. dataset (N=823) 8 and the Banchereau et al. dataset (N=247) 54 . The Braun et al. data consists in primary site tumors and metastases of advanced ccRCC samples from patients that progressed on 1, 2 or 3 previous therapies (at least one systemic anti-angiogenic therapy). These patients were included in clinical trials CM-009 ( NCT01358721 ), CM-010 ( NCT01354431 ) and CM-025 ( NCT01668784 ) treated with anti-PD-1 antibody nivolumab or mTOR inhibitor everolimus (EGAC00001001519; EGAC00001001520; EGAC00001001521). The Beuselinck et al. data consists in 53 primary site ccRCC samples treated with sunitinib. The Zhang et al. data of 94 ccRCC primary site tumor samples from patient treated with sunitinib. To be consistent with the other datasets, we selected only 51 samples with stages III-IV from the Zhang et al. dataset. The Motzer et al. data consists in 823 primary or metastatic samples of ccRCC treated with sunitinib or atezolizumab+bevacizumab. The Banchereau et al. RCC data consists in 247 samples of ccRCC treated with sunitinib, atezolizumab+bevacizumab or atezolizumab alone. Clustering of samples based on methyl-array data ccRCC samples were clustered according beta values from methyl-array data. The Arai et al. method divided ccRCC samples by hierarchical clustering (agglomeration method = Ward, distance = “euclidean”) into a hyper-methylated profile named CIMP+ (noted “High” in figures) and the remaining CIMP-(“Low”) based on beta values from 801 probes selected on >27,000 probe values obtained via 27K Illumina array. The Sato et al. method split ccRCC samples by hierarchical clustering (agglomeration method = Ward.D2, distance = “euclidean”) into three subtypes (high, intermediate, low) based on beta values from 1668 probes selected on >450,000 probes values obtained via 450K Illumina array. The Evelonn et al. method divided ccRCC samples by a classification score based on 172 probes. Following the merging of different platforms and missing values filtering, only a subset of these signatures was used to cluster ccRCC samples according to the Arai et al. method (531 out of 801 probes), Sato et al. method (1006 out of 1672 probes) and Evelönn method (148 out of 174 probes). The Lu et al., method divided ccRCC samples into 2 methylation subtypes based on 58 probes. We used the iMES package to perform the clustering (citation here: https://github.com/xlucpu/iMES/tree/main ) (Lu et al., 2023). Because the method could not be used with missing values, we imputed by k-means (k=10) 5 genes filtered out by the missing value processing. A summary of all probe signatures processing could be found Figure 1 . Gene expression data analyses Consensus Clustering of k-means was used to cluster ccRCC samples into two transcriptomic subtypes (parameters: maxK=6, reps = 100, pItem = 0.8, pFeature = 0.8, clusterAlg = “km”, distance = “euclidean”). Two-sided Wilcoxon rank-sum tests were used to compare gene expression between hyper and hypo-methylated samples in the TCGA and Sato et al. datasets. The 20,085 overlapping genes between both datasets were tested. Cellular deconvolution method implemented in CIBERSORTx (version 1.0, 12/21/2019) 71 was applied to predict tumor micro-environment (TME) composition from bulk gene gene expression data. The CIBERSORTx cell fractions module was executed using the docker image after registration and access token received ( https://cibersortx.stanford.edu/download.php ). A refined and relabeled single-cell RNA-seq data was used as ccRCC reference 37 , 72 . Single-sample Gene Set Enrichment Analysis (ssGSEA) of the oncogenic signatures (C6) and the hallmark (H) gene sets from the MSigDB database (Human MSigDB v2023.1.Hs) 38 , 39 was performed using the R/Bioconductor package GSVA (v. 1.42.0). Activity of transcription factors (TFs) was inferred from gene expression of known targets for each given TF contained in the DoRothEA database from the DoRothEA R/Bioconductor package (v. 1.12.0) 41 . Thereby, 271 TF-target collections were retained from the DoRothEA A, B, and C interaction confidence levels. Only TFs with at least 15 measured transcripts were included. For the activity inference, the VIPER algorithm 40 was used based on the log2-transformed fold-change of gene expression values between Hyper- and Hypo-methylated profiles from the Sato et al. method. The VIPER analysis was carried out by the msviper() function (parameters: minsize = 15, adaptive.size = F, pleiotropy = TRUE) from the viper R/Bioconductor package (v. 1.28.0). Cellular deconvolution, ssGSEA and TF activity inference analyzes were performed separately on each cohort. Cell fractions from cellular deconvolution and ssGSEA values were separately standardized by features for each cohort (Z-score). Only significant cell fractions, ssGSEA pathways and TFs in all of the three cohorts were selected for further analysis. P values from ssGSEA and TF activity analyzes from each cohort were merged by the Stouffer’s pooling method. NES values from the three cohorts of TF activities were averaging for visualization purposes. All the intermediate files could be found in the supplementary tables. A tertiary lymphoid structure (TLS) score was calculated, based on a 7-gene signature (CCL19, CCL21, CXCL13, CCR7, CXCR5, SELL, LAMP3), as previously described 34 . TLS values were divided by their median values into high and low status for visualization purposes. Raw score values were used for statistical comparison between methylation subtypes (two-sided Wilcoxon rank-sum test). Correlation analysis between DNA methylation levels and gene expression values To compare the DNA methylation levels and gene expression values of matching methyl-array and gene expression data used in our gene expression-based analyzes (TCGA, CPTAC, Sato et al.), we used probe annotations from the IlluminaHumanMethylation450kanno.ilmn12.hg19 R/Bioconductor package (v. 0.6.0). We then retained probes that were both associated to the genes of interest and annotated as “Promoter_Associated” from the “Regulatory_Feature_Group” feature. For each gene, we averaged their probes values in case of multiple probes by gene. Statistical analyses Statistical tests Statistical difference of numerical continuous values between two groups was assessed using two-sided Wilcoxon rank-sum tests with the Benjamini-Hochberg correction for multiple hypothesis testing. For proportion comparisons, Pearson’s chi-squared test was used when the number of samples was greater than 5 for each group, otherwise, one- or two-sided Fisher’s exact tests were used. Survival analysis Survival analyses were performed using the R package survival (version 3.4-0). Survival curves of progression-free survival (PFS) and overall survival (OS) were plotted using the Kaplan–Meier method. The statistical comparison of the survival outcomes between groups was performed by log-rank tests from the R package survminer (version 0.4.9). We defined as free of recurrence the patients without progression in PFS information and with a follow-up greater than 6 months. Patients with a follow-up lower than 6 months and without a recurrence was removed for the cancer recurrence analysis. Classification analyses For the gene ratio-based analyses, the inverted ratios with the same performances as the reversed that is retained were removed for visualization purposes. For AUC-ROC values, the hyper-methylated group (“High” in figures) was retained as the positive class. Therefore, features were marked of a symbol “(-)” when high values were associated with the negative class (“Low” in figures, the hypo-methylated group). The best threshold for classification obtained on the TCGA dataset was found by the Youden’s index method with the cutpointr() function from the cutpointr R package (v. 1.1.2). The CPTAC dataset was used as a independent dataset to validate each threshold expressed in transcripts per million (TPM) values from RNA-Seq data. Funding The study was supported by the KATY project, which has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No 101017453 and by the CANVAS project, which has received funding from the Horizon Europe twinning program under grant agreement No 101079510, and by the DIGPHAT project (Multi-scale and longitudinal data modeling in pharmacology: toward digital pharmacological twins), which has received funding from the French research initiative “France 2030” through the program PEPR Digital Health under ANR grant agreement No 22-PESN-0017. DP’s research is supported by Agence Nationale de la Recherche under projects ProFI (Proteomics French Infrastructure, ANR-10-INBS-08) and GRAL, a program from the Chemistry Biology Health (CBH) Graduate School of University Grenoble Alpes (ANR-17-EURE-0003). Author contributions Conceptualization: FJ, CB, DP Methodology: FJ, CB, DP Investigation: FJ, SS Visualization: FJ Supervision: CB, DP Writing—original draft: FJ, CB, DP Writing—review & editing: FJ, CB, DP, PD, HA, SNS, AL, SS Competing interests Authors declare that they have no competing interests. Data and materials availability The TCGAbiolinks R package (v. 2.22.4) was used to download TCGA and CPTAC datasets (methyl-array and RNA-Seq). The Sato et al. datasets were collected from ArrayExpress (E-MTAB-2007). The Beuselinck et al. datasets were collected from ArrayExpress (E-MTAB-3267). The Evelönn et al., Braun et al. and Zhang et al. datasets were collected from the data sources of the articles. The Motzer et al. normalized gene expression data (TPM) generated by RNA-Seq and clinical data of 823 participants were obtained from the European Genome-phenome Archive (EGA) (EGAC00001001813). The Banchereau et al. normalized gene expression data (TPM) generated by RNA-Seq and clinical data were obtained from the EGA (EGAC00001001748). Acknowledgments Most of the computations presented in this paper were performed using the GRICAD infrastructure ( https://gricad.univ-grenoble-alpes.fr ), which is supported by Grenoble research communities. We acknowledge Genentech for the access to Motzer et al. (EGAC00001001813) and Banchereau et al. (EGAC00001001748) datasets of normalized gene expression data and associated clinical features References 1. ↵ Ljungberg , B. et al. European Association of Urology Guidelines on Renal Cell Carcinoma: The 2022 Update . European Urology 82 , 399 – 410 ( 2022 ). OpenUrl CrossRef PubMed 2. ↵ Hsieh , J. J. et al. Renal cell carcinoma . Nat Rev Dis Primers 3 , 1 – 19 ( 2017 ). OpenUrl 3. ↵ Atkins , M. B. & Tannir , N. M. Current and emerging therapies for first-line treatment of metastatic clear cell renal cell carcinoma . Cancer Treatment Reviews 70 , 127 – 137 ( 2018 ). OpenUrl CrossRef PubMed 4. ↵ Yang , J. , Wang , K. & Yang , Z. Treatment strategies for clear cell renal cell carcinoma: Past, present and future . Frontiers in Oncology 13 , ( 2023 ). 5. ↵ Motzer , R. J. et al. Nivolumab for Metastatic Renal Cell Carcinoma: Results of a Randomized Phase II Trial . JCO 33 , 1430 – 1437 ( 2015 ). OpenUrl Abstract / FREE Full Text 6. ↵ Motzer , R. J. et al. Nivolumab plus Ipilimumab versus Sunitinib in Advanced Renal-Cell Carcinoma . New England Journal of Medicine 378 , 1277 – 1290 ( 2018 ). OpenUrl CrossRef PubMed 7. ↵ Rini , B. I. et al. Pembrolizumab plus Axitinib versus Sunitinib for Advanced Renal-Cell Carcinoma . New England Journal of Medicine 380 , 1116 – 1127 ( 2019 ). OpenUrl CrossRef PubMed 8. ↵ Motzer , R. J. et al. Molecular Subsets in Renal Cancer Determine Outcome to Checkpoint and Angiogenesis Blockade . Cancer Cell 38 , 803 – 817 .e4 ( 2020 ). OpenUrl CrossRef PubMed 9. ↵ Braun , D. A. et al. Interplay of somatic alterations and immune infiltration modulates response to PD-1 blockade in advanced clear cell renal cell carcinoma . Nat Med 26 , 909 – 918 ( 2020 ). OpenUrl CrossRef PubMed 10. ↵ Morad , G. , Helmink , B. A. , Sharma , P. & Wargo , J. A. Hallmarks of response, resistance, and toxicity to immune checkpoint blockade . Cell 184 , 5309 – 5337 ( 2021 ). OpenUrl CrossRef PubMed 11. ↵ Creighton , C. J. et al. Comprehensive molecular characterization of clear cell renal cell carcinoma . Nature 499 , 43 – 49 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 12. ↵ Sato , Y. et al. Integrated molecular analysis of clear-cell renal cell carcinoma . Nat Genet 45 , 860 – 867 ( 2013 ). OpenUrl CrossRef PubMed 13. ↵ Clark , D. J. et al. Integrated Proteogenomic Characterization of Clear Cell Renal Cell Carcinoma . Cell 179 , 964 – 983 .e31 ( 2019 ). OpenUrl CrossRef PubMed 14. ↵ Li , Y. et al. Histopathologic and proteogenomic heterogeneity reveals features of clear cell renal cell carcinoma aggressiveness . Cancer Cell 41 , 139 – 163 .e17 ( 2023 ). OpenUrl CrossRef PubMed 15. ↵ Zhang , H. et al. Proteogenomics of clear cell renal cell carcinoma response to tyrosine kinase inhibitor . Nat Commun 14 , 4274 ( 2023 ). OpenUrl CrossRef PubMed 16. ↵ Davis , A. A. & Patel , V. G. The role of PD-L1 expression as a predictive biomarker: an analysis of all US Food and Drug Administration (FDA) approvals of immune checkpoint inhibitors . J Immunother Cancer 7 , 278 ( 2019 ). OpenUrl Abstract / FREE Full Text 17. ↵ Shum , B. , Larkin , J. & Turajlic , S. Predictive biomarkers for response to immune checkpoint inhibition . Seminars in Cancer Biology 79 , 4 – 17 ( 2022 ). OpenUrl CrossRef PubMed 18. ↵ Wood , M. A. , Weeder , B. R. , David , J. K. , Nellore , A. & Thompson , R. F. Burden of tumor mutations, neoepitopes, and other variants are weak predictors of cancer immunotherapy response and overall survival . Genome Medicine 12 , 33 ( 2020 ). OpenUrl CrossRef PubMed 19. McGrail , D. J. et al. High tumor mutation burden fails to predict immune checkpoint blockade response across all cancer types . Annals of Oncology 32 , 661 – 672 ( 2021 ). OpenUrl CrossRef PubMed 20. ↵ Motzer , R. J. et al. Avelumab plus axitinib versus sunitinib in advanced renal cell carcinoma: biomarker analysis of the phase 3 JAVELIN Renal 101 trial . Nat Med 26 , 1733 – 1741 ( 2020 ). OpenUrl CrossRef PubMed 21. ↵ Wu , C.-C. , Wang , Y. A. , Livingston , J. A. , Zhang , J. & Futreal , P. A. Prediction of biomarkers and therapeutic combinations for anti-PD-1 immunotherapy using the global gene network association . Nat Commun 13 , 42 ( 2022 ). OpenUrl CrossRef PubMed 22. ↵ Ayers , M. et al. IFN-γ–related mRNA profile predicts clinical response to PD-1 blockade . J Clin Invest 127 , 2930 – 2940 ( 2017 ). OpenUrl CrossRef PubMed 23. ↵ Liu , D. et al. Integrative molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma . Nat Med 25 , 1916 – 1927 ( 2019 ). OpenUrl CrossRef PubMed 24. ↵ Auslander , N. et al. Robust prediction of response to immune checkpoint blockade therapy in metastatic melanoma . Nat Med 24 , 1545 – 1549 ( 2018 ). OpenUrl CrossRef PubMed 25. ↵ Lei , Y. , Li , X. , Huang , Q. , Zheng , X. & Liu , M. Progress and Challenges of Predictive Biomarkers for Immune Checkpoint Blockade . Frontiers in Oncology 11 , ( 2021 ). 26. ↵ Rosellini , M. et al. Prognostic and predictive biomarkers for immunotherapy in advanced renal cell carcinoma . Nat Rev Urol 1 – 25 ( 2022 ) doi: 10.1038/s41585-022-00676-0 . OpenUrl CrossRef PubMed 27. ↵ Arai , E. et al. Single-CpG-resolution methylome analysis identifies clinicopathologically aggressive CpG island methylator phenotype clear cell renal cell carcinomas . Carcinogenesis 33 , 1487 – 1493 ( 2012 ). OpenUrl CrossRef PubMed 28. ↵ Evelönn , E. A. et al. DNA methylation associates with survival in non-metastatic clear cell renal cell carcinoma . BMC Cancer 19 , 65 ( 2019 ). OpenUrl CrossRef PubMed 29. Wei , J.-H. et al. A CpG-methylation-based assay to predict survival in clear cell renal cell carcinoma . Nat Commun 6 , 8699 ( 2015 ). OpenUrl CrossRef PubMed 30. ↵ El Khoury , L. Y. et al. Identification of DNA methylation signatures associated with poor outcome in lower-risk Stage, Size, Grade and Necrosis (SSIGN) score clear cell renal cell cancer . Clinical Epigenetics 13 , 12 ( 2021 ). OpenUrl CrossRef PubMed 31. ↵ Duruisseaux , M. et al. Epigenetic prediction of response to anti-PD-1 treatment in non-small-cell lung cancer: a multicentre, retrospective analysis . The Lancet Respiratory Medicine 6 , 771 – 781 ( 2018 ). OpenUrl CrossRef PubMed 32. Filipski , K. et al. DNA methylation-based prediction of response to immune checkpoint inhibition in metastatic melanoma . J Immunother Cancer 9 , e002226 ( 2021 ). OpenUrl Abstract / FREE Full Text 33. ↵ Lu , X. et al. Silencing of genes by promoter hypermethylation shapes tumor microenvironment and resistance to immunotherapy in clear-cell renal cell carcinomas . Cell Reports Medicine 4 , 101287 ( 2023 ). OpenUrl CrossRef PubMed 34. ↵ Dieu-Nosjean , M.-C. , Goc , J. , Giraldo , N. A. , Sautès-Fridman , C. & Fridman , W. H. Tertiary lymphoid structures in cancer and beyond . Trends in Immunology 35 , 571 – 580 ( 2014 ). OpenUrl CrossRef PubMed 35. Petitprez , F. , Meylan , M. , de Reyniès , A. , Sautès-Fridman , C. & Fridman , W. H. The Tumor Microenvironment in the Response to Immune Checkpoint Blockade Therapies . Frontiers in Immunology 11 , ( 2020 ). 36. ↵ Simonaggio , A. et al. Tumor Microenvironment Features as Predictive Biomarkers of Response to Immune Checkpoint Inhibitors (ICI) in Metastatic Clear Cell Renal Cell Carcinoma (mccRCC) . Cancers (Basel) 13 , 231 ( 2021 ). OpenUrl PubMed 37. ↵ Jeanneret , F. et al. Microenvironment of metastatic site reveals key predictors of PD-1 blockade response in renal cell carcinoma . 2023.07.17.548676 Preprint at doi: 10.1101/2023.07.17.548676 ( 2023 ). OpenUrl Abstract / FREE Full Text 38. ↵ Liberzon , A. et al. Molecular signatures database (MSigDB) 3.0 . Bioinformatics 27 , 1739 – 1740 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 39. ↵ Liberzon , A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection . Cell Syst 1 , 417 – 425 ( 2015 ). OpenUrl CrossRef PubMed 40. ↵ Alvarez , M. J. et al. [VIPER]Functional characterization of somatic mutations in cancer using network-based inference of protein activity . Nat Genet 48 , 838 – 847 ( 2016 ). OpenUrl CrossRef PubMed 41. ↵ Garcia-Alonso , L. , Holland , C. H. , Ibrahim , M. M. , Turei , D. & Saez-Rodriguez , J. Benchmark and integration of resources for the estimation of human transcription factor activities . Genome Res . 29 , 1363 – 1375 ( 2019 ). OpenUrl Abstract / FREE Full Text 42. ↵ Morgan , R. & Pandha , H. S. PBX3 in Cancer . Cancers (Basel) 12 , 431 ( 2020 ). OpenUrl PubMed 43. Du , F. et al. SOX13 promotes colorectal cancer metastasis by transactivating SNAI2 and c-MET . Oncogene 39 , 3522 – 3540 ( 2020 ). OpenUrl CrossRef PubMed 44. ↵ Vasefifar , P. et al. Nanog, as a key cancer stem cell marker in tumor progression . Gene 827 , 146448 ( 2022 ). OpenUrl CrossRef PubMed 45. ↵ Mellor , P. et al. CREB3L1 Is a Metastasis Suppressor That Represses Expression of Genes Regulating Metastasis, Invasion, and Angiogenesis . Molecular and Cellular Biology 33 , 4985 – 4995 ( 2013 ). OpenUrl Abstract / FREE Full Text 46. ↵ Zheng , R. & Blobel , G. A. GATA Transcription Factors and Cancer . Genes Cancer 1 , 1178 – 1188 ( 2010 ). OpenUrl CrossRef PubMed 47. Chou , J. , Provot , S. & Werb , Z. GATA3 in development and cancer differentiation: Cells GATA have it! Journal of Cellular Physiology 222 , 42 – 49 ( 2010 ). OpenUrl CrossRef PubMed 48. Shi , Q. et al. GATA3 suppresses human fibroblasts-induced metastasis of clear cell renal cell carcinoma via an anti-IL6/STAT3 mechanism . Cancer Gene Ther 27 , 726 – 738 ( 2020 ). OpenUrl CrossRef PubMed 49. ↵ Peters , I. et al. Decreased mRNA expression of GATA1 and GATA2 is associated with tumor aggressiveness and poor outcome in clear cell renal cell carcinoma . Target Oncol 10 , 267 – 275 ( 2015 ). OpenUrl CrossRef PubMed 50. ↵ Gordon , G. J. et al. Using Gene Expression Ratios to Predict Outcome Among Patients With Mesothelioma . JNCI: Journal of the National Cancer Institute 95 , 598 – 605 ( 2003 ). OpenUrl CrossRef PubMed Web of Science 51. Ma , X.-J. et al. A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen . Cancer Cell 5 , 607 – 616 ( 2004 ). OpenUrl CrossRef PubMed Web of Science 52. ↵ Reddy , A. et al. Gene Expression Ratios Lead to Accurate and Translatable Predictors of DR5 Agonism across Multiple Tumor Lineages . PLOS ONE 10 , e0138486 ( 2015 ). OpenUrl CrossRef PubMed 53. ↵ Beuselinck , B. et al. Molecular Subtypes of Clear Cell Renal Cell Carcinoma Are Associated with Sunitinib Response in the Metastatic Setting . Clinical Cancer Research 21 , 1329 – 1339 ( 2015 ). OpenUrl Abstract / FREE Full Text 54. ↵ Banchereau , R. et al. Molecular determinants of response to PD-L1 blockade across tumor types . Nat Commun 12 , 3969 ( 2021 ). OpenUrl CrossRef PubMed 55. ↵ Motzer , R. J. , Bacik , J. , Murphy , B. A. , Russo , P. & Mazumdar , M. Interferon-Alfa as a Comparative Treatment for Clinical Trials of New Therapies Against Advanced Renal Cell Carcinoma . JCO 20 , 289 – 296 ( 2002 ). OpenUrl Abstract / FREE Full Text 56. ↵ Neidhart , M. Neidhart , M. Chapter 7 - DNA Methylation and Cancer . in DNA Methylation and Complex Human Disease (ed. Neidhart , M. ) 103 – 134 ( Academic Press, Oxford , 2016 ). doi: 10.1016/B978-0-12-420194-1.00007-5 . OpenUrl CrossRef 57. ↵ Rasmussen , K. D. & Helin , K. Role of TET enzymes in DNA methylation, development, and cancer . Genes Dev 30 , 733 – 750 ( 2016 ). OpenUrl Abstract / FREE Full Text 58. ↵ Dowsett , M. et al. Assessment of Ki67 in Breast Cancer: Recommendations from the International Ki67 in Breast Cancer Working Group . JNCI: Journal of the National Cancer Institute 103 , 1656 – 1664 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 59. ↵ Zheng , D. , Ning , J. , Xia , Y. , Ruan , Y. & Cheng , F. Comprehensive analysis of a homeobox family gene signature in clear cell renal cell carcinoma with regard to prognosis and immune significance . Front Oncol 12 , 1008714 ( 2022 ). OpenUrl CrossRef PubMed 60. ↵ Madden , E. , Logue , S. E. , Healy , S. J. , Manie , S. & Samali , A. The role of the unfolded protein response in cancer progression: From oncogenesis to chemoresistance . Biology of the Cell 111 , 1 – 17 ( 2019 ). OpenUrl CrossRef PubMed 61. ↵ Hsu , S.-K. et al. Unfolded Protein Response (UPR) in Survival, Dormancy, Immunosuppression, Metastasis, and Treatments of Cancer Cells . International Journal of Molecular Sciences 20 , 2518 ( 2019 ). OpenUrl CrossRef PubMed 62. ↵ Lv , L. et al. IGF2BP3 prevent HMGB1 mRNA decay in bladder cancer and development . Cellular & Molecular Biology Letters 29 , 39 ( 2024 ). OpenUrl CrossRef PubMed 63. Ueki , A. et al. Up-Regulation of Imp3 Confers In Vivo Tumorigenicity on Murine Osteosarcoma Cells . PLOS ONE 7 , e50621 ( 2012 ). OpenUrl CrossRef PubMed 64. ↵ Panebianco , F. et al. THADA fusion is a mechanism of IGF2BP3 activation and IGF1R signaling in thyroid cancer . Proceedings of the National Academy of Sciences 114 , 2307 – 2312 ( 2017 ). OpenUrl Abstract / FREE Full Text 65. ↵ Chen , C.-L. et al. Reciprocal regulation by TLR4 and TGF-β in tumor-initiating stem-like cells . J Clin Invest 123 , 2832 – 2849 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 66. ↵ Li , M. et al. An isocorydine derivative (d-ICD) inhibits drug resistance by downregulating IGF2BP3 expression in hepatocellular carcinoma . Oncotarget 6 , 25149 – 25160 ( 2015 ). OpenUrl CrossRef PubMed 67. ↵ Mancarella , C. & Scotlandi , K. IGF2BP3 From Physiology to Cancer: Novel Discoveries, Unsolved Issues, and Future Perspectives . Front Cell Dev Biol 7 , 363 ( 2020 ). OpenUrl CrossRef PubMed 68. ↵ Liu , X. et al. Targeting IGF2BP3 in Cancer . Int J Mol Sci 24 , 9423 ( 2023 ). OpenUrl CrossRef PubMed 69. ↵ Feng , Y. et al. Insulin-like growth factor-2 mRNA-binding protein 3 promotes cell migration, invasion, and epithelial−mesenchymal transition of esophageal squamous cell carcinoma cells by targeting zinc finger E-box-binding homeobox 1 mRNA . Molecular Carcinogenesis 62 , 503 – 516 ( 2023 ). OpenUrl CrossRef PubMed 70. ↵ Schmit , K. & Michiels , C. TMEM Proteins in Cancer: A Review . Frontiers in Pharmacology 9 , ( 2018 ). 71. ↵ Newman , A. M. et al. Determining cell type abundance and expression from bulk tissues with digital cytometry . Nat Biotechnol 37 , 773 – 782 ( 2019 ). OpenUrl CrossRef PubMed 72. ↵ Obradovic , A. et al. Single-cell protein activity analysis identifies recurrence-associated renal tumor macrophages . Cell 184 , 2988 – 3005 .e16 ( 2021 ). OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted November 06, 2024. Download PDF Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following DNA methylation is a key determinant of response to targeted and immune checkpoint therapies in metastatic renal cell carcinoma Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share DNA methylation is a key determinant of response to targeted and immune checkpoint therapies in metastatic renal cell carcinoma Florian Jeanneret , Sarah Schoch , Pedro Ballester , Stefan N Symeonides , Alexander Laird , Håkan Axelson , Delphine Pflieger , Christophe Battail bioRxiv 2024.11.05.622095; doi: https://doi.org/10.1101/2024.11.05.622095 Share This Article: Copy Citation Tools DNA methylation is a key determinant of response to targeted and immune checkpoint therapies in metastatic renal cell carcinoma Florian Jeanneret , Sarah Schoch , Pedro Ballester , Stefan N Symeonides , Alexander Laird , Håkan Axelson , Delphine Pflieger , Christophe Battail bioRxiv 2024.11.05.622095; doi: https://doi.org/10.1101/2024.11.05.622095 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Bioinformatics Subject Areas All Articles Animal Behavior and Cognition (7651) Biochemistry (17746) Bioengineering (13928) Bioinformatics (42066) Biophysics (21499) Cancer Biology (18650) Cell Biology (25579) Clinical Trials (138) Developmental Biology (13409) Ecology (19947) Epidemiology (2067) Evolutionary Biology (24374) Genetics (15633) Genomics (22557) Immunology (17775) Microbiology (40505) Molecular Biology (17217) Neuroscience (88796) Paleontology (667) Pathology (2845) Pharmacology and Toxicology (4836) Physiology (7664) Plant Biology (15179) Scientific Communication and Education (2047) Synthetic Biology (4304) Systems Biology (9839) Zoology (2272)